Scoring the Data Using Association Rules

摘要: In many data mining applications, the objective is to select cases of a target class. For example, in direct marketing, marketers want likely buyers particular product for promotion. such it often too difficult predict who will definitely be class (e.g., buyer class) because used modeling very noisy and has highly imbalanced distribution. Traditionally, classification systems are solve this problem. Instead classifying each case definite or non-buyer), system modified produce probability estimate (or score) indicate likelihood that belongs class). However, existing only aim find subset regularities rules exist data. This gives partial picture domain. paper, we show selection problem can mapped association rule provide more powerful solution Since aims all data, thus able give complete underlying relationships The set enables us assign accurate case. paper proposes an effective efficient technique compute estimates using rules. Experiment results public domain real-life application general new performs markedly better than state-of-the-art C4.5, boosted Naive Bayesian system.